NOS protein conformation modeling.

Although our computational approach to predicting NOS conformation is conceptually similar to the protein folding calculations (see, e.g., Ref. [¹⁴] for a recent review), the focus of our calculations is different. While the protein folding calculations aim at predicting a single most probable tertiary structure, our calculations here generate a statistical distribution of all possible conformations of the random coil fragments joining the NOS domains. Acquiring a sufficient conformational statistics requires a large amount of calculations, which necessitates the simplifying assumptions employed in our structural modeling below.

One has to note that the calculations of protein conformations using the distance restraints obtained by pulsed EPR are in principle possible with the aid of the open source Matlab-based software for multiscale macromolecular modeling (MMM).^15–16 However, MMM handles the protein residues explicitly and is therefore not sufficiently fast for our purposes. This and multiple other aspects of the problem (e.g., two heme centers interacting with the BSL, necessity to incorporate docking probability restraints, etc.) have prevented us from using MMM and motivated us develop the software specifically suited for our needs.

The Monte Carlo modeling of NOS conformational distribution was done using a home-written Delphi program. When possible, the random numbers used in these calculations were generated by the hardware random number generator available on the Intel processor chips starting from Ivy Bridge and on the AMD processors starting 2015 (the RDRAND function). If an older computer was used in the calculations, then the random numbers were generated by the high quality pseudo random number generator, Mersenne Twister MT19937,¹⁷ seeded from the system clock. The essential features of the conformational modeling are discussed in the Results and Discussion section.